MozReview-Commit-ID: Lrxi8t53nwy
If a developer adds '--rebuild N' to their try syntax they will get test jobs scheduled N times.
This is useful to determine intermittency rate.
This fixes a regression due to the recent refactoring on how we schedule tasks.
--HG--
extra : rebase_source : 355ca631353015bf63461c194168d753efd6958e
I was looking at some decision task logs and noticed lines like:
Connection pool is full, discarding connection: taskcluster
I also noticed lines like:
Starting new HTTP connection (153): taskcluster
In this case, we had established 153 TCP connections to a server.
Looking at the requests source code, a requests.session by default
creates a connection pool with capacity of 10. There are actually 2
components to the capacity: idle connections and active connections.
What appeared to be happening was we could obtain an idle connection,
use it, and then it would be discarded when put back in the idle
pool because the idle pool was at capacity. Furthermore, it also
appears that connections were sitting around waiting for a TCP
connection.
This commit uses a custom "adapter" with an increased pool size that
matches the concurrency level of the code issuing the HTTP requests.
This should increase the number of concurrent TCP connections / requests,
decreease the number of TCP connections being used overall, and make
decision tasks complete faster.
MozReview-Commit-ID: 6NDbz78TM2y
--HG--
extra : rebase_source : a48e1dae31a5e2af2bc554b544a9019d50dc784f
This introduces a completely new way of specifying test task in-tree,
completely replacing the old spider-web of YAML files.
The high-level view is this:
- some configuration files are used to determine which test suites to run
for each test platform, and against which build platforms
- each test suite is then represented by a dictionary, and modified by a
sequence of transforms, duplicating as necessary (e.g., chunks), until
it becomes a task definition
The transforms allow sufficient generality to support just about any desired
configuration, with the advantage that common configurations are "easy" while
unusual configurations are supported but notable for their oddness (they
require a custom transform).
As of this commit, this system produces the same set of test graphs as the
existing YAML, modulo:
- extra.treeherder.groupName -- this was not consistent in the YAML
- extra.treeherder.build -- this is ignored by taskcluster-treeherder anyway
- mozharness command argument order
- boolean True values for environment variables are now the string "true"
- metadata -- this is now much more consistent, with task name being the label
Testing of this commit demonstrates that it produces the same set of test tasks for
the following projects (those which had special cases defined in the YAML):
- autoland
- ash (*)
- willow
- mozilla-inbound
- mozilla-central
- try:
-b do -p all -t all -u all
-b d -p linux64,linux64-asan -u reftest -t none
-b d -p linux64,linux64-asan -u reftest[x64] -t none[x64]
(*) this patch omits the linux64/debug tc-M-e10s(dt) test, which is enabled on
ash; ash will require a small changeset to re-enable this test.
IGNORE BAD COMMIT MESSAGES (because the hook flags try syntax!)
MozReview-Commit-ID: G34dg9f17Hq
--HG--
rename : taskcluster/taskgraph/kind/base.py => taskcluster/taskgraph/task/base.py
rename : taskcluster/taskgraph/kind/docker_image.py => taskcluster/taskgraph/task/docker_image.py
rename : taskcluster/taskgraph/kind/legacy.py => taskcluster/taskgraph/task/legacy.py
extra : rebase_source : 03e70902c2d3a297eb9e3ce852f8737c2550d5a6
extra : histedit_source : d4d9f4b192605af21f41d83495fc3c923759c3cb
Currently, Gecko decision tasks spend ~8 minutes making HTTP requests to the
queue API. Let's throw a thread pool at it so tasks are submitted faster.
This could possibly be faster if there were a single batch submit API in
the queue. However, that isn't implemented for technical reasons described
in the bug.
MozReview-Commit-ID: MeHItEVBbk
--HG--
extra : rebase_source : ebe182c5906df70cd2e4cbd545f7b7b9189882da
* Implement & document optimization (although legacy kind doesn't do much of it)
* Introduce `optimize_target_tasks` parameter to control whether tasks in the
target set can be optimized (no for try, yes for most other branches)
* Refactor to include resolved taskIds in the optimized task graph
* Include a `label-to-taskid.json` artifact.
* Introduce {'task-reference': '... <dependency-name> ...'} for referring to
parent tasks' taskId.
MozReview-Commit-ID: LWvlWNz49U5
--HG--
extra : rebase_source : 780e0e23d24b268ade33ecdcbccb5081f32aac48
The `taskgraph` package generates TaskCluster task graphs based on collections
of task "kinds". Initially, there is only one kind, the "legacy" kind, which
reads the YAML files from `testing/taskcluster/tasks` to generate the task
graph.
Try syntax is implemented by filtering the tasks in the taskgraph after it has
been created, then extending the result to include any prerequisite tasks.
A collection of `mach taskgraph` subcommands are provided for developers to
extend or debug the task-graph generation process.
MozReview-Commit-ID: 1TJCns4XxZ8
--HG--
rename : testing/taskcluster/docs/index.rst => taskcluster/docs/index.rst
extra : rebase_source : 7b9125281d66044db9bd8e4a1fade16136f384b9
extra : histedit_source : 47640d27080acda0279270babbcf33f5badb0d1c
